Which description best captures testing sensitivity to missing data?

Which description best captures testing sensitivity to missing data?

• A. Exploring how results change under different plausible imputations or bounds
• B. Deleting all incomplete cases
• C. Assuming missing data has no effect
• D. Collecting new data only after policy decisions

Answer: (A)

Testing sensitivity to missing data asks how conclusions change when missing values are handled in different plausible ways. The best description is to explore results under different imputations or bounds for the missing data. This approach, a sensitivity analysis, shows whether findings are robust to the assumptions about why data are missing and how they’re filled in. If results stay similar across reasonable imputations or bounds, you have more confidence in the conclusions; if they shift, the missing data mechanism is materially influencing the results. Deleting incomplete cases ignores information that missingness may carry, and can bias results. Assuming missing data has no effect pretends there’s no potential bias, which isn’t a genuine sensitivity test. Collecting new data after policies are decided doesn’t address how existing results would respond to different missing-data assumptions.

Testing sensitivity to missing data asks how conclusions change when missing values are handled in different plausible ways. The best description is to explore results under different imputations or bounds for the missing data. This approach, a sensitivity analysis, shows whether findings are robust to the assumptions about why data are missing and how they’re filled in. If results stay similar across reasonable imputations or bounds, you have more confidence in the conclusions; if they shift, the missing data mechanism is materially influencing the results. Deleting incomplete cases ignores information that missingness may carry, and can bias results. Assuming missing data has no effect pretends there’s no potential bias, which isn’t a genuine sensitivity test. Collecting new data after policies are decided doesn’t address how existing results would respond to different missing-data assumptions.